与LinkedBlockingQueue相比，Apache Ignite Queue慢得多

Question

我正在尝试在Ignite中复制一个简单的生产者-消费者方案：

public class QueueExample {
    public static void main(String[] args) {
        new QueueExample().start();
    }

    private void start() {
        final AtomicBoolean finishedTest1 = new AtomicBoolean(false);
        final BlockingQueue<Double> queue = new LinkedBlockingQueue<>(5);
        final CountDownLatch latch = new CountDownLatch(2);
        final int MAX = 1000;

        new Thread(() -> {
            System.out.println("test1 before latch");
            latch.countDown();
            try {
                // wait until other runnable is able to poll
                latch.await(20, TimeUnit.SECONDS);
            } catch (Exception ex) {
                throw new RuntimeException(ex);
            }
            System.out.println(new Date().getTime() + " start test1");
            double test = 2;
            Random r = new Random();
            StopWatch sw = new StopWatch();
            sw.start();
            for (int i = 0; i < MAX; i++) {
                try {
                    queue.put(r.nextDouble());
                } catch (InterruptedException e) {
                    throw new RuntimeException(e);
                }
            }
            sw.stop();
            finishedTest1.set(true);
            //LoggerFactory.getLogger(getClass()).info
            System.out.println(new Date().getTime() + " end test1. " + test + ", took:" + sw.getTime() / 1000f);
        }).start();

        new Thread(() -> {
            System.out.println("test2 before latch");
            latch.countDown();
            try {
                // wait until other runnable is able to poll
                latch.await(10, TimeUnit.SECONDS);
            } catch (Exception ex) {
                throw new RuntimeException(ex);
            }
            System.out.println(new Date().getTime() + " start test2");
            StopWatch sw = new StopWatch();
            sw.start();
            int counter = 0;
            try {
                for (int i = 0; i < MAX ; i++) {
                    Double res = queue.poll(1, TimeUnit.SECONDS);
                    counter++;
                }
            } catch (InterruptedException e) {
                // expected
            }
            sw.stop();

            //LoggerFactory.getLogger(getClass()).info
            System.out.println(new Date().getTime() + " end test2. counter " + counter + ", finished:" + finishedTest1.get() + ", took:" + sw.getTime() / 1000f);
        }).start();
    }
}

为什么与下面的Ignite代码相比，这快100倍（0.02秒与<2秒）？

public class MyIgnite {
    public static void main(String[] args) {
        new MyIgnite().start();
    }

    private void start() {
        IgniteConfiguration icfg = new IgniteConfiguration();
        icfg.setIgniteInstanceName("test1");
        Ignite ignite1 = Ignition.start(icfg);

        final CountDownLatch latch = new CountDownLatch(2);

        final int queueSize = 5;
        CollectionConfiguration queueCfg = new CollectionConfiguration();

        ignite1.compute().runAsync(new IgniteRunnable() {

            @IgniteInstanceResource
            Ignite ignite;

            @Override
            public void run() {
                IgniteQueue<Double> queue = ignite.queue("test", queueSize, queueCfg);
                System.out.println("test1 fetched queue");
                latch.countDown();
                try {
                    // wait until other runnable is able to poll
                    latch.await(20, TimeUnit.SECONDS);
                } catch (Exception ex) {
                    throw new RuntimeException(ex);
                }
                System.out.println("start test1");
                double test = 2;
                Random r = new Random();
                StopWatch sw = new StopWatch();
                sw.start();
                for (int i = 0; i < 1000; i++) {
                    queue.put(r.nextDouble());
                }
                sw.stop();
                //LoggerFactory.getLogger(getClass()).info
                System.out.println("end test1. " + test + " at ignite " + ignite.name() + ", took:" + sw.getTime() / 1000f);
            }
        });

        System.out.println("starting test2");
        icfg = new IgniteConfiguration();
        icfg.setIgniteInstanceName("test2");
        Ignite ignite2 = Ignition.start(icfg);
        ignite2.compute().runAsync(new IgniteRunnable() {
            @IgniteInstanceResource
            Ignite ignite;

            @Override
            public void run() {
                IgniteQueue<Double> queue = ignite.queue("test", queueSize, queueCfg);
                System.out.println("test2 fetched queue");
                latch.countDown();
                try {
                    // wait until other runnable is able to poll
                    latch.await(10, TimeUnit.SECONDS);
                } catch (Exception ex) {
                    throw new RuntimeException(ex);
                }
                System.out.println("start test2");
                StopWatch sw = new StopWatch();
                sw.start();
                int counter = 0;
                try {
                    for (int i = 0; i < 1000; i++) {
                        Double res = queue.poll(5, TimeUnit.SECONDS);
                        counter++;
                    }

                } catch (IgniteException exc) {
                    System.out.println("Somehow cannot poll. " + exc);
                }
                sw.stop();
                //LoggerFactory.getLogger(getClass()).info
                System.out.println("end test2. counter " + counter + " at ignite " + ignite.name() + ", took:" + sw.getTime() / 1000f);
            }
        });

        System.out.println("oldest node: " + ignite1.cluster().forOldest().hostNames());
        System.out.println("nodes: " + ignite1.cluster().nodes().size());

        // does it really gracefully shut the nodes down?
//        Ignition.stop(ignite1.name(), false);
//        Ignition.stop(ignite2.name(), false);
    }
}

我试图修改点火存储，使其行为更像内存中的，但是更改数字失败。

Answer 1

您正在将航空母舰与玩具船进行比较。

LinkedBlockingQueue是在单个JVM的内存中工作的数据结构。

IgniteQueue是基于Ignite键值存储的分布式结构。它可以在数百台具有不同一致性级别的计算机上工作，并具有持久性。当然，它由许多内部机制支持，并且比简单的本地队列要慢。